Task-adversarial co-generative nets

Pei Yang, Hanghang Tong, Qi Tan, Jingrui He

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we propose Task-Adversarial co-Generative Nets (TAGN) for learning from multiple tasks. It aims to address the two fundamental issues of multi-task learning, i.e., domain shift and limited labeled data, in a principled way. To this end, TAGN first learns the task-invariant representations of features to bridge the domain shift among tasks. Based on the task-invariant features, TAGN generates the plausible examples for each task to tackle the data scarcity issue. In TAGN, we leverage multiple game players to gradually improve the quality of the co-generation of features and examples by using an adversarial strategy. It simultaneously learns the marginal distribution of task-invariant features across different tasks and the joint distributions of examples with labels for each task. The theoretical study shows the desired results: at the equilibrium point of the multi-player game, the feature extractor exactly produces the task-invariant features for different tasks, while both the generator and the classifier perfectly replicate the joint distribution for each task. The experimental results on the benchmark data sets demonstrate the effectiveness of the proposed approach.

Original languageEnglish (US)
Title of host publicationKDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
PublisherAssociation for Computing Machinery
Pages1596-1604
Number of pages9
ISBN (Electronic)9781450362016
DOIs
StatePublished - Jul 25 2019
Externally publishedYes
Event25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019 - Anchorage, United States
Duration: Aug 4 2019Aug 8 2019

Publication series

NameProceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Conference

Conference25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019
CountryUnited States
CityAnchorage
Period8/4/198/8/19

Fingerprint

Labels
Classifiers

Keywords

  • Generative adversarial nets
  • Multi-task learning

ASJC Scopus subject areas

  • Software
  • Information Systems

Cite this

Yang, P., Tong, H., Tan, Q., & He, J. (2019). Task-adversarial co-generative nets. In KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1596-1604). (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). Association for Computing Machinery. https://doi.org/10.1145/3292500.3330843

Task-adversarial co-generative nets. / Yang, Pei; Tong, Hanghang; Tan, Qi; He, Jingrui.

KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 2019. p. 1596-1604 (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Yang, P, Tong, H, Tan, Q & He, J 2019, Task-adversarial co-generative nets. in KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Association for Computing Machinery, pp. 1596-1604, 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019, Anchorage, United States, 8/4/19. https://doi.org/10.1145/3292500.3330843
Yang P, Tong H, Tan Q, He J. Task-adversarial co-generative nets. In KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery. 2019. p. 1596-1604. (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). https://doi.org/10.1145/3292500.3330843
Yang, Pei ; Tong, Hanghang ; Tan, Qi ; He, Jingrui. / Task-adversarial co-generative nets. KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 2019. pp. 1596-1604 (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining).
@inproceedings{f2be5e69773d447fa414cfe03960d8fd,
title = "Task-adversarial co-generative nets",
abstract = "In this paper, we propose Task-Adversarial co-Generative Nets (TAGN) for learning from multiple tasks. It aims to address the two fundamental issues of multi-task learning, i.e., domain shift and limited labeled data, in a principled way. To this end, TAGN first learns the task-invariant representations of features to bridge the domain shift among tasks. Based on the task-invariant features, TAGN generates the plausible examples for each task to tackle the data scarcity issue. In TAGN, we leverage multiple game players to gradually improve the quality of the co-generation of features and examples by using an adversarial strategy. It simultaneously learns the marginal distribution of task-invariant features across different tasks and the joint distributions of examples with labels for each task. The theoretical study shows the desired results: at the equilibrium point of the multi-player game, the feature extractor exactly produces the task-invariant features for different tasks, while both the generator and the classifier perfectly replicate the joint distribution for each task. The experimental results on the benchmark data sets demonstrate the effectiveness of the proposed approach.",
keywords = "Generative adversarial nets, Multi-task learning",
author = "Pei Yang and Hanghang Tong and Qi Tan and Jingrui He",
year = "2019",
month = "7",
day = "25",
doi = "10.1145/3292500.3330843",
language = "English (US)",
series = "Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",
publisher = "Association for Computing Machinery",
pages = "1596--1604",
booktitle = "KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",

}

TY - GEN

T1 - Task-adversarial co-generative nets

AU - Yang, Pei

AU - Tong, Hanghang

AU - Tan, Qi

AU - He, Jingrui

PY - 2019/7/25

Y1 - 2019/7/25

N2 - In this paper, we propose Task-Adversarial co-Generative Nets (TAGN) for learning from multiple tasks. It aims to address the two fundamental issues of multi-task learning, i.e., domain shift and limited labeled data, in a principled way. To this end, TAGN first learns the task-invariant representations of features to bridge the domain shift among tasks. Based on the task-invariant features, TAGN generates the plausible examples for each task to tackle the data scarcity issue. In TAGN, we leverage multiple game players to gradually improve the quality of the co-generation of features and examples by using an adversarial strategy. It simultaneously learns the marginal distribution of task-invariant features across different tasks and the joint distributions of examples with labels for each task. The theoretical study shows the desired results: at the equilibrium point of the multi-player game, the feature extractor exactly produces the task-invariant features for different tasks, while both the generator and the classifier perfectly replicate the joint distribution for each task. The experimental results on the benchmark data sets demonstrate the effectiveness of the proposed approach.

AB - In this paper, we propose Task-Adversarial co-Generative Nets (TAGN) for learning from multiple tasks. It aims to address the two fundamental issues of multi-task learning, i.e., domain shift and limited labeled data, in a principled way. To this end, TAGN first learns the task-invariant representations of features to bridge the domain shift among tasks. Based on the task-invariant features, TAGN generates the plausible examples for each task to tackle the data scarcity issue. In TAGN, we leverage multiple game players to gradually improve the quality of the co-generation of features and examples by using an adversarial strategy. It simultaneously learns the marginal distribution of task-invariant features across different tasks and the joint distributions of examples with labels for each task. The theoretical study shows the desired results: at the equilibrium point of the multi-player game, the feature extractor exactly produces the task-invariant features for different tasks, while both the generator and the classifier perfectly replicate the joint distribution for each task. The experimental results on the benchmark data sets demonstrate the effectiveness of the proposed approach.

KW - Generative adversarial nets

KW - Multi-task learning

UR - http://www.scopus.com/inward/record.url?scp=85071184994&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85071184994&partnerID=8YFLogxK

U2 - 10.1145/3292500.3330843

DO - 10.1145/3292500.3330843

M3 - Conference contribution

AN - SCOPUS:85071184994

T3 - Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

SP - 1596

EP - 1604

BT - KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

PB - Association for Computing Machinery

ER -